Fix flaky alerts going dormant for high-recovery-volume cases

Fixed flaky-test alerts “wedging” silent for high-volume recovery cases. Rolling, recovery-enabled flaky alerts that quarantine many tests were silently stopping all muting, marking, and recovery actions because the recovery query overflowed ClickHouse limits (the evaluation cursor would freeze and the reprocessing window would grow forever). The query now aggregates counts inside ClickHouse, honors each candidate’s own trigger cutoff, and processes candidates in batches so it stays within request limits. Affected alerts self-heal on their next tick, and chronically flaky tests get re-muted while they remain flaky.